has revealed details about some important techniques it will use to make computer chips in the future, advances seen as essential for the company to continue its steady progress toward faster, smaller, cheaper microprocessors.
In a wide-ranging speech, executives at the Intel Developer Forum described plans to use new materials and structures to create the microscopic transistors that cover the surface of silicon chips. Besides boosting PC performance, the techniques should enable chips that combine communications with computation, reducing the cost and boosting the power of mobile phones, network equipment and other gear.
"We envision a future where information becomes more personal, where my communications device is in my lapel pin, where I have instant access to knowledge whatever I am doing and wherever I am," said Pat Gelsinger, Intel vice president and chief technology officer.
Transistors act like tiny switches on the surface of chips, turning on and off at lightning speed to represent the ones and zeros of binary computer code. As transistors get smaller and multiply in number, however, challenges arise that threaten to block the progress of Moore's Law, a decades-old prediction that the number of transistors on a chip will double roughly every two years.
One challenge being addressed by Intel, IBM and other manufacturers is finding a way to pass more electrical current through transistors without them burning up or leaking electricity. IBM has announced plans to introduce a "double gate" transistor around 2006, which should allow it to increase electrical current and thus the performance of its chips.
Not to be outdone, Intel revealed plans to build what it called "Tri-Gate" transistors soon after the middle of the decade. By manipulating structures on its chips that are only as wide as a few dozen atoms, the company has devised a way to increase the surface area of each transistor gate, producing the equivalent of three gates for each transistor.
Intel plans to present more details about the design, along with "record-setting performance data," at a technical conference in Japan next week, said Sunlin Chou, senior vice president and general manager of Intel's technology and manufacturing group.
Time will tell which method proves most successful, said Nathan Brookwood, principal analyst with US firm, Insight 64. Both efforts are in their early stages and each company could potentially end up using a technique being developed by the other, he said. The important thing for users is that chipmakers are investing to overcome such technological barriers.
Intel also revealed that it will use a new material, Silicon Germanium, to make some of its chips when it moves to a new "90 nanometer" manufacturing process toward the end of next year. The material, already in use by IBM, is a better conductor of electricity and can help boost chip performance. It is also more expensive, Chou said, and Intel will restrict its use initially to chips used in communications equipment such as optical networking components.
Boosting the transistor count will allow Intel to integrate new functionality on its processors. For example, by the middle of the decade it expects to combine analogue components like a radio frequency receiver with logic components, creating in effect a miniature and inexpensive radio on a single chip. For users it means that any device with an Intel processor will in effect include wireless capabilities for free.
New components for networking gear are also in Intel’s sights. Gelsinger, the company’s chief technology officer, described efforts to build a "tunable laser" — an important component for optical networks that today costs thousands of dollars. By combining digital and optical functions on a single chip, and by making the chips using standard manufacturing processes, Intel expects to produce such components later in the decade for as little as a few dollars, he said.
The advance could make it far less expensive to "light up" the miles of optical fibre that currently lie unused around the world, boosting the amount of bandwidth available to end users, said analyst Brookwood.